You are on page 1of 9

70 SMART MATERIALS, MICROGELS Vol.

STATISTICAL THERMODYNAMICS
Introduction

The objective and mode of thinking of statistical thermodynamics, its elementary


tools, and model building are directly relevant to polymer physics. This article
introduces the subject and relevant key references from a modern point of view.
Perhaps the best introductory material is the problem source book (1,2), while
Reference 3 is a standard reference. Minimum prerequisites are the rudiments of
thermodynamics (4,5), and mechanics (6,7). Understanding of probability is very
helpful (8).

Encyclopedia of Polymer Science and Technology. Copyright John Wiley & Sons, Inc. All rights reserved.
Vol. 8 STATISTICAL THERMODYNAMICS 71

Objective and Mode of Thinking

Macroscopic objects contain many atoms and molecules. Although these elements
may obey the laws of (quantum) mechanics, it is extremely costly, and often mean-
ingless, to know the mechanical motion of each element. Fortunately, these objects
obey their own macroscopic laws, as demonstrated empirically. Typical examples
of these laws are thermodynamics and hydrodynamics. These laws describe rela-
tions among macroscopic observables. They are usually called phenomenological
laws and are often regarded as less fundamental than, eg, the laws of mechanics.
However, the laws of, eg, thermodynamics are asymptotically exact, and should
not be regarded as an inaccurate approximation of something more fundamental.
It should not be forgotten that the laws of mechanics are also empirical laws. The
founding fathers of statistical mechanics, including Gibbs and Einstein, tried to
justify thermodynamics with mechanics, but the subsequent history tells us that
the mechanics they relied on was less reliable than thermodynamics.
To understand systems consisting of many elements with the aid of thermo-
dynamics and mechanics is the objective of statistical thermodynamics. The ad-
vent of single macromolecule experimental methods emphasizes the importance
of the study of mesoscopic systems and fluctuations. Macroscopic and mesoscopic
observables (henceforth simply called macro-observables) are interpreted as sums
of many microscopic quantities or coarse-grained stochastic quantities. Thus, elu-
cidating macroscopic or mesoscopic behavior in terms of the properties of elements
inevitably becomes statistical. The laws of large numbers and the large deviation
principle (8) become the key mathematical tools.
The problems in statistical thermodynamics fall into two categories. The first
category involves the study of the structure of phenomenological framework and
the interrelations among macro-observables without calculating their actual val-
ues. The second category involves the calculation of the actual values of param-
eters appearing in the phenomenology from more microscopic parameters. The
first category concerns finding and explaining the relationships among macro-
observables appearing in the description of a class of states of materials. The
two main problems are to recognize well-defined classes of phenomena, called
universality classes, and to construct suitable models (the so-called minimal mod-
els) for these classes. These models are statistical but their elements need not
be molecules or atoms. There is a wide range of choice of elementary objects;
truly atomistic pictures can even be counterproductive. A typical problem in the
second category is to compute phenomenological parameters, eg, viscosities or
phase-transition temperatures, which, by definition, reflect the specificity of the
individual systems. For the second-category problems, the source of the micro-
scopic parameters is delicate; they are often obtainable only through macroscopic
comparisons. Thus, it is difficult to obtain model-free conclusions, although prac-
tically useful results may be obtained.

Equilibrium Problems

The main theoretical framework of the static aspect of statistical thermodynam-


ics is Gibbs statistical mechanics. The universal behavior of macroscopic objects
72 STATISTICAL THERMODYNAMICS Vol. 8

can be described phenomenologically with the aid of equilibrium thermodynam-


ics with five major laws [(4,5); those who are serious about thermodynamics
must read Ref. 9]: the existence of thermal equilibrium and temperature (the
zeroth law); the conservation of energy (the first law); the variational principle
for entropy (the second law); the inaccessibility of the absolute zero temperature
(the third law); and the clear distinction between extensive and intensive quan-
tities (sometimes called the fourth law). The equilibrium states of macroscopic
objects comprise a universality class characterized by equilibrium thermodynam-
ics, and statistical ensembles should be understood as minimal models for this
class.
All macroscopic observables are obtainable from the distribution of micro-
scopic states (henceforth, microstates) of elements that obey mechanics. Strictly
speaking, this is the most basic assumption of statistical thermodynamics. How-
ever, to elucidate macroscopic phenomena, it is not necessary to know the true
distribution of microstates of the system. Boltzmann introduced the concept of
orthodic ensembles that are compatible with thermodynamics. In practice, an
orthodic ensemble is established by demonstrating that it is compatible with the
laws of mechanics and with the laws of thermodynamics. Gibbs demonstrated that
the canonical distribution function

1 − H({q})/kB T
P({q}) = e (1)
Z

with the micro–macro relation dictated by the expression for the Helmholtz free
energy A:

A= − kB T log Z (2)

satisfies the above compatibility requirement. Here, {q} collectively denotes the
microscopic coordinates, H({q}) is the Hamiltonian of the system, kB is the Boltz-
mann constant, and T is the absolute temperature. Z is the normalization constant
of the distribution called the canonical partition function:

Z= e − H/kB T (3)

where the summation is the sum (integral) over all the microscopic states. A phys-
ical derivation of the canonical ensemble was given by Einstein, whose starting
point was the principle of equal probability: the probability of finding a microstate
on a constant-energy surface is independent of the microstate (the microcanoni-
cal ensemble). Einstein introduced the canonical ensemble as a subensemble of
a microcanonical ensemble. Then, he identified heat as the energy imported to
the microscopic degrees of freedom, and derived equation 2 (for a historical ac-
count see Ref. 10). There are many arguments demonstrating the naturalness of
the canonical distribution. Lenard (11) demonstrated, crudely put, that the dis-
tribution of microstates as given by the canonical distribution is equivalent to
Vol. 8 STATISTICAL THERMODYNAMICS 73

the second law in the form of “passivity”, A ≤ W, where W is the work done to
the system to change it from an initial to a final equilibrium state at the same
temperature, with A being the free-energy change between these two states. A
stronger statement is possible [Jarzynski’s equality (12)]
 
e − W/kB T = e − A/kB T (4)

Here, A = Ab − Aa , where As is the Helmholtz free energy of the equilibrium


state with the Hamiltonian H s (s = a or b) with the temperature T, and   implies
the average over many experiments, where W is the work in each run of the
experiment changing the Hamiltonian from H a to H b according to a fixed protocol
with the initial ensemble given by the canonical distribution ∝ e − Ha /kB T . Although
it is not easy to use this equality for a macroscopic system, it is practically useful
for single-molecule experiments (cf. 13).
Mathematically speaking, statistical mechanics must study a system in the
thermodynamic (large system) limit. In this limit, the system Hamiltonian does
not exist. Thus, we need an indirect way to introduce the distribution (measure)
in the thermodynamic limit. This is the theory of the Gibbs measure (14,15).
If we wish to understand phase transitions as mathematical singularities of
thermodynamic potentials, we must take the thermodynamic limit. When the
Gibbs measure for a system is not unique the system exhibits a phase transition.
If we look at a small portion of a macroscopic system or study a mesoscopic
system, we must study fluctuations. The probability of fluctuations is phenomeno-
logically described by the thermodynamic theory of fluctuations (5). From the en-
semble theory point of view, the fluctuation theory is the study of large deviations
from the expectation value. This is the reason why large deviation theory is be-
coming increasingly important in statistical thermodynamics. Standard works on
large deviation theory are References 16 and 17; perhaps as accessible introduc-
tion to the topic may be found in Reference 18.

Nonequilibrium Problems

Even more challenging problems than those discussed above arise from the con-
sideration of time-dependent and nonequilibrium properties of macroscopic sys-
tems. Unfortunately, there is no general theoretical framework tantamount to the
Gibbs statistical mechanics. Most likely, there is no universality class that can
be called the macroscopic nonequilibrium class. There is, however, a class that
may be called the linear nonequilibrium class. Its phenomenological framework is
the linear nonequilibrium thermodynamics, which contains several fundamental
laws: the second law as the positivity of entropy produced inside the system as a re-
sult of dissipation; phenomenological linear force–flux relations; and the Onsager
reciprocity stating the symmetry relations among kinetic coefficients (19,20). We
also assume that thermodynamic relations hold locally in space (this is called the
local equilibrium assumption).
The fundamental statistical thermodynamic idea for the linear nonequilib-
rium class is that fluctuations can spontaneously realize nonequilibrium states in
74 STATISTICAL THERMODYNAMICS Vol. 8

a (small) part of a large equilibrium system. A precise formulation was given as


Onsager’s regression hypothesis: the decay dynamics of fluctuations is on the av-
erage the macroscopic laws governing the time evolution of macro-observables. At
least formally, for example, the Navier–Stokes equation (and the associated heat
equation) may be derived as the equation governing the fluctuation of the velocity
field (21). The statistical principle is given as Onsager’s principle [so christened by
Hashitsume (22)] that can be summarized, in modern terms, roughly as follows:
the Lagrangian of the Onsager–Machlup path integral is given by the large devi-
ation function for the microscopic fluctuations. This framework tells us that the
response of an observable of a system to a small external disturbance is controlled
by the fluctuation dynamics of the observable [fluctuation–dissipation relations
(FDR) (23,24)]. The FDR of the first kind relates the flux–flux correlation function
to the corresponding transport coefficient (the Green–Kubo relations), and the
FDR of the second kind gives the relation between the autocorrelation of the noise
and the mobility (25). The formula for the intrinsic viscosity in terms of polymer
force flux may be a familiar example for polymer scientists (26). The FDR of the
second kind is of supreme importance in the construction of minimal models to
elucidate time-dependent properties of systems (27) close to equilibrium. Although
FDR is often called the fluctuation–dissipation theorem, its derivation is always
a combination of microscopic argument and phenomenological requirements, as
clearly stressed by Nakano (24).
Needless to say, most interesting phenomena in nonequilibrium are not lin-
ear. These phenomena cannot be in the class considered above, but usually they are
modeled in terms of nonlinear Langevin equations of the following form (Zwanzig’s
nonlinear Langevin equation):

dA δF
= v(A) − L + f (5)
dt δA

or the corresponding Fokker-Planck equation describing the time evolution of the


distribution function (28–30). Here, v(A) is called the streaming term, L is the
(bare Onsager) kinetic coefficient (matrix), F is the free-energy functional, and
f is the noise satisfying the FDR of the second kind: f f(t) = 2kB TLδ(t). It is
assumed that the equilibrium distribution of A is described by the distribution
function proportional to exp(−F/kB T). The streaming term v in equation 5 is
often expanded in terms of the polynomials of A. When this expansion is trun-
cated at the second order, equation 5 becomes the starting point of the mode-
coupling theory and dynamical renormalization group theory, with a remarkable
success in critical dynamics (31). The use of the mode-coupling theory for su-
percooled liquids and glasses seems to be an increasingly important topic with
highly interesting results (32), but its microscopic foundation is still a difficult
question.
It seems that all the successfully elucidated dynamical universality classes
adopt as their minimal models the ones described in terms of nonlinear Langevin
equations (33). This indicates the general correctness of Onsager’s principle as
a fundamental principle (beyond the linear regime). Just as equilibrium sta-
tistical mechanics is a statistical framework based on the principle (of equal
Vol. 8 STATISTICAL THERMODYNAMICS 75

probability) compatible with the phenomenological laws (equilibrium thermo-


dynamics), nonequilibrium statistical mechanics exhibits the same structure.
Thus, to deepen our understanding of nonequilibrium phenomena and to go
beyond the linear regime, we need clear phenomenological understanding of a
wider class of phenomena. There are some attempts (34,35), but there is no
consensus.

Synopsis and Examples

Statistical thermodynamics is a discipline that elucidates phenomenological laws


of a class of phenomena (a universality class) from analyses of statistical models
that capture the salient features of the class (minimal models). The principal
objective of statistical thermodynamics is to recognize universality classes and
to find corresponding minimal models. Examples follows from polymer solution
theory.
Dilute Solution. If the interchain interactions can be ignored, the poly-
mer solution is in the dilute regime. If we are interested in the global features
of a polymer, we need not respect the discreteness of each bond; we may use the
Wiener process to describe the chain [thanks to Donsker’s invariance principle
(36)]. The interaction among nonbonded monomer units gives the well-known
excluded-volume interaction and is usually mimicked by a delta-function pseu-
dopotential. Thus, a chain described by the Wiener process interacting with it-
self through a delta-function pseudopotential (with an interaction contour-range
cutoff) is conceived as a minimal model for static properties of dilute solutions
[the Edwards model (37)]. The partition function of a chain can be conveniently
described in terms of the path integral (the Feynman–Kac formula) (38,39). RG
theory (for general introduction, see Refs. 40–43) has given various quantitative
results for universal properties; agreement with experiment looks satisfactory
(44–46). That is, we can clearly recognize a universality class of dilute-solution
statics with a good minimal model.
To understand dynamics and transport properties, a set of (nonlinear)
Langevin equations describing the Edwards model and the solvent velocity
field is proposed as a minimal model. Comparison of the outcome of the model
with experimental results has revealed that this minimal model is reason-
able (47); but closely studied, less than minimal. The importance of the di-
rect chain–chain friction must not be underestimated (48). With this addi-
tional effect, the model seems minimal, but a more careful study may be
needed.
Semidilute Systems. In more concentrated solutions, the interchain in-
teractions (overlap) cannot be ignored. Even if the monomer number density of
the solution is infinitesimal, long chains can overlap extensively. This regime is
the semidilute solution recognized by des Cloizeaux. In this regime the inter-
chain interactions may still be described by the delta-function pseudopotential.
For equilibrium properties the many-chain version of Edward’s minimal model
seems appropriate (49,50). Thus, we have the universality class of semidilute solu-
tion statics with a reliable minimal model. For dynamics, since there is extensive
chain overlap, the chain entanglement cannot be ignored. The straightforward
76 STATISTICAL THERMODYNAMICS Vol. 8

many-body version of the dilute-solution minimal Langevin model is intractable


and it is fair to say that our understanding of semidilute-solution dynamics is
incomplete (51). As a minimal model, a model of fluids with the strain field ef-
fect by Doi and Onuki should be studied seriously (33). To understand dynamics
in polymer solutions in general, careful simulations would become increasingly
important (52).
Beyond Semidilute Solutions. Further increase of concentration gives
the concentrated regime and the melt regime. Beyond the semidilute regime, the
minimal model mentioned above is not adequate, because the volume fraction is
now finite and it really exerts entropic forces (53). If the static model is not in a
good shape, it should be clear that we do not have a reliable minimal model for
dynamics of nondilute solutions. We must take into account entanglement and
direct friction as well as the hydrodynamic interactions. Thus, from the statistical
thermodynamic point of view, real understanding of many polymer chain systems
is still beyond our reach.

BIBLIOGRAPHY

“Statistical Thermodynamics” in EPSE 2nd ed., Vol. 15, pp. 614–625, by Y. Oono, University
of Illinois.
1. M. Toda, R. Kubo, and N. Saito, Statistical Physics I Equilibrium Statistical Mechanics,
Springer-Verlag, New York, 1983.
2. R. Kubo, H. Ichimaru, T. Usui, and N. Hashitsume, Statistical Mechanics, North Hol-
land, Amsterdam, 1965.
3. L. D. Landau and E. M. Lifshitz, Statistical Mechanics, Pergamon Press, London,
1980.
4. R. Kubo, H. Ichimaru, T. Usui, and N. Hashitsume, Thermodynamics, North Holland,
Amsterdam, 1968.
5. H. B. Callen, Thermodynamics, Interscience Publishers, New York, 1960.
6. L. D. Landau and E. M. Lifshitz, Mechanics, Pergamon Press, London, 1976.
7. G. A. Baym, Lectures on Quantum Mechanics, Benjamin/Cummings Publishing Co.,
Reading, Mass., 1973.
8. R. Durrett, Probability: Theory and Examples, Wadsworth & Brooks Cole, Pacific Grove,
Calif., 1991.
9. E. H. Lieb and J. Yngvason, Phys. Rep. 310, 1 (1999).
10. L. Navarro, Arch. Hist. Exact Sci. 53, 147 (1998).
11. A. Lenard, J. Stat. Phys. 19, 575 (1978).
12. C. Jarzynski, Phys. Rev. Lett. 78, 2690 (1997).
13. J. Liphardt, S. Dumont, S. B. Smith, I. Tinoco Jr., and C. Bustamante, Science 296,
1832 (2002).
14. D. Ruelle, Statistical Mechanics, Rigorous Results, Benjamin, New York,
1969.
15. R. B. Israel, Convexity in the Theory of Lattice Gases, Princeton University Press,
Princeton, 1979.
16. J.-D. Deuschel and D. W. Strook, Large Deviation, Academic Press, New York,
1989.
17. R. S. Ellis, Entropy, Large Deviations, and Statistical Mechanics, Springer-Verlag, New
York, 1985.
18. Y. Oono, Prog. Theor. Phys. Suppl. 99, 165 (1989).
Vol. 8 STATISTICAL THERMODYNAMICS 77

19. I. Prigogine, Introduction to Thermodynamics of Irreversible Processes, 2nd ed., Wiley-


Interscience, New York, 1961.
20. S. R. de Groot and P. Mazur, Nonequilibrium Thermodynamics, North Holland, Ams-
terdam, 1969 [A Dover edition is also available].
21. K. Miyazaki, K. Kitahara, and D. Bedeaux, Physica A 230, 600 (1996).
22. N. Hashitsume, Prog. Theor. Phys. 8, 461 (1952).
23. R. Kubo, Rep. Prog. Phys. 29, 255 (1966).
24. H. Nakano, Int. J. Mod. Phys. B 7, 2397 (1993).
25. H. Mori, Prog. Theor. Phys. 33, 424 (1965).
26. M. Bixon, Annu. Rev. Phys. Chem. 27, 65 (1976).
27. B. J. Berne, ed., Statistical Mechanics, Part B: Time-Dependent Processes, Plenum
Press, New York, 1977.
28. H. Risken, The Fokker–Planck Equation, 2nd ed., Springer-Verlag, Berlin,
1989.
29. C. W. Gardiner, Handbook of Stochastic Methods for Physics, Chemistry and the Natural
Sciences, 2nd ed., Springer-Verlag, Berlin, 1985.
30. N. G. van Kampen, Stochastic Processes in Physics and Chemistry, North Holland,
1981.
31. K. Kawasaki, Ann. Phys. 61, 1 (1970).
32. W. Götze and L. Sjögren, Rep. Prog. Phys. 55, 241 (1992).
33. A. Onuki, Phase Transition Dynamics, Cambridge University Press, Cambridge,
2002.
34. D. Jou, J. Casas-Vásquez, and G. Lebon, Extended Irreversible Thermodynamics,
Springer-Verlag, New York, 1993.
35. Y. Oono and M. Paniconi, Prog. Theor. Phys. Suppl. 130, 27 (1998).
36. M. D. Donsker, Mem. Am. Math. Soc. 6 (1951).
37. S. F. Edwards, Proc. Phys. Soc. 85, 613 (1965).
38. R. P. Feynman, Statistical Mechanics, Benjamin/Cummings Publishing, Co., Reading,
Mass., 1972.
39. I. M. Gel’fand and A. M. Yaglom, J. Math. Phys. 1, 48 (1960).
40. M. E. Fisher, in F. J. W. Hahne, ed., Lecture Notes in Physics, Vol. 186, Springer-Verlag,
New York, 1983.
41. M. Le Bellac, Quantum and Statistical Field Theory, Oxford University Press, Oxford,
1991.
42. N. D. Goldenfeld, Lectures on Phase Transitions and the Renormalization Group,
Addison-Wesley Publishing Co., Inc., Reading, Mass., 1992.
43. Y. Oono, Advances in Chemical Physics, Vol. 61, Wiley-Interscience, New York, 1985,
Chapt. 5.
44. P. G. de Gennes, Scaling Concepts in Polymer Physics, Cornell University Press, Ithaca,
N.Y., 1980.
45. M. Daoud, J. P. Cotton, B. Farnoux, G. Jannink, G. Sarma, H. Benoit, R. Duplessix, C.
Picot, and P. G. de Gennes, Macromolecules 8, 804 (1975).
46. J. des Cloizeaux and G. Jannink, Polymers in Solution: Their Modelling and Structure,
Oxford University Press, Oxford, 1990.
47. W. Brown and T. Nicolai, in W. Brown, ed., Dynamic Light Scattering, Oxford University
Press, Oxford, 1993, p. 272.
48. Y. Shiwa and Y. Tsunashima, Physica A 197, 47 (1993).
49. T. Ohta and A. Nakanishi, J. Phys. A 16, 4155 (1983); T. Ohta and Y. Oono, Phys. Lett.
A 89, 460 (1982).
50. L. Schäfer, Excluded Volume Effects in Polymer Solutions, as Explained by the Renor-
malization Group, Springer, Berlin, 1999.
51. G. C. Berry, Adv. Polym. Sci. 114, 233 (1994).
78 STATISTICAL THERMODYNAMICS Vol. 8

52. B. Dünweg, D. Reith, M. Steinhauser, and K. Kremer, J. Chem. Phys. 117, 914
(2002); P. Ahlrichs, R. Evraers, and B. Dünweg, Phys. Rev. E 64, 040501(R)
(2001).
53. Y. Shiwa, Y. Oono, and P. W. Baldwin, Mod. Phys. Lett. B 4, 1421 (1990).

YOSHITSUGU OONO
University of Illinois

STRESS RELAXATION. See VISCOELASTICITY.

You might also like